Train re_identification_net with more than 10 cameras using TAO

[alexander103]

Hello, I am training a re_identification_net model using Tao V5.2.0. But I have a dataset of 11 cameras. When I train the model using the notebook reidentificationnet_resnet.ipynb the model never sees images with names like 0001_c10s1_000004.jpg or 0001_c11s1_000113.jpg I think due to the use of the pattern re. compile(r'([-d]+)_c(\d)'), which only expects a one-digit pattern after the letter ‘c’. Is there a parameter I can modify so that the training inside the container accepts more than 10 cameras?

[Morganh]

@alexander103

the pattern re. compile(r'([-d]+)_c(\d)')

For workaround, you can modify tao_pytorch_backend/nvidia_tao_pytorch/cv/re_identification/model/pl_reid_model.py at 9c2d94c0635b1117edfea85a94a6e3d0ead53754 · NVIDIA/tao_pytorch_backend · GitHub. This pl_reid_model.py can be found inside the docker.

[alexander103]

Great, thanks!!! What would be the correct way to modify the pl_reid_model.py script, and update the called container when we are using TAO through CLI on the notebook?

!tao model re_identification train \
                  -e $SPECS_DIR/experiment_market1501_resnet.yaml \
                  -r $RESULTS_DIR/$exp_path \
                  -k $KEY

[Morganh]

You can open a terminal and run docker run.
$ docker run --runtime=nvidia -it --rm-v /local-folder:/docker-folder nvcr.io/nvidia/tao/tao-toolkit:5.3.0-pyt /bin/bash
Then inside the docker,
$ mv /usr/local/lib/python3.10/dist-packages/nvidia_tao_pytorch/cv/re_identification/model/pl_reid_model.py /usr/local/lib/python3.10/dist-packages/nvidia_tao_pytorch/cv/re_identification/model/pl_reid_model.py.bak
$ vim /usr/local/lib/python3.10/dist-packages/nvidia_tao_pytorch/cv/re_identification/model/pl_reid_model.py (copy the content from tao_pytorch_backend/nvidia_tao_pytorch/cv/re_identification/model/pl_reid_model.py at 9c2d94c0635b1117edfea85a94a6e3d0ead53754 · NVIDIA/tao_pytorch_backend · GitHub

Then still run training inside the docker.
$ re_identification train xxx

[yingliu]

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks